Revolutionizing AI Conversations: OpenAI's New Update to ChatGPT Tackles the 'AI Laziness' Challenge

AI-driven chatbots are software or computer programs powered by artificial intelligence (AI) to simulate human conversations. Chatbots utilize machine learning and Natural Language Processing (NLP) algorithms to analyze and respond to real-time user inquiries.

We can’t underestimate the use cases of AI-driven chatbots. You can use AI chatbots like ChatGPT for Data Analysis, code generation, creating conversational interfaces, and content generation, among other use cases.

Despite the growing use of these chatbots, it has been argued that AI is becoming lazy. A tweet published on the official ChatGPT X account confirms that the team at OpenAI received reports that GPT4 is getting lazier.

In this context, “AI Laziness” refers to a situation where AI responses, while factually correct, often lack depth and contextual understanding. For instance, you can request ChatGPT to generate code for you, but it stops halfway. In other cases, it forgets the initial conversation.

This has been a concern, but OpenAI, one of the industry leaders, has a solution.

OpenAI update

The observation by the users and response by the OpenAI team indicated that something needed to be fixed. OpenAI published a blog post on January 25, 2024, introducing new updates that address the “AI Laziness” issue, improve content moderation, and introduce new pricing models.

The update's major focus is on the following:

• An updated GPT-4 Turbo preview model
• An updated GPT-3.5 Turbo model

The gpt-4-0125-preview focuses on code generation, where users said that ChatGPT was underperforming. The gpt-4-0125 is still in preview mode (in development mode and undergoing testing). According to the post, GPT-4 Turbo will launch in a few months.

The update introduces “embeddings,” a sophisticated AI feature that enhances how the model analyzes inputs and generates outputs. There have also been reports of bugs in non-English UTF-8 generations, which the gpt-4-0125-preview addresses.

Users can now expect higher accuracy in responses requested in various formats through the updated GPT-3.5 Turbo model. Token prices will also be $0.0005 /1K tokens for inputs and $0.0015 /1K for outputs.

How Embeddings Work

The update will likely influence How ChatGPT Works, as the embeds are designed to improve the LLMs. Embeddings are sequences of numbers representing the concepts within the content. The concepts in question here can be code or natural language.

Embeddings parse relationships between content, facilitating complex tasks like retrieval and clustering. Such embeds, thus, power uses like content retrieval in ChatGPT and its APIs.

New pricing model

Existing AI companies must ensure they price their products competitively as new companies and models continue to arise. OpenAI has restructured the pricing of GPT-3.5 Turbo in this update. The gpt-3.5-turbo-0125 has slashed output prices by 25%, while inputs are now 50% lower.

The update seeks to strike a balance between affordability and scaling. You will be automatically upgraded to gpt-3.5-turbo-0125 if you have already subscribed to GPT-3.5 Turbo.

Technical Insights of the update

The recent update by OpenAI introduces three embeddings to its LLMs. According to the company, Text-embedding-3-large is the best-performing of the three embeds. The company claims that tests show benchmark scores have risen from 31% to 54% to 61 to 64%.

Some experts like Dwayne Charrington, an Aurelia.ai core member, have already supported the claims. In a tweet, Dwayne says, “I don’t know what you did, but you fixed it”. He goes on to say that you still have to prompt the GPT correctly.

In the press release, OpenAI explains, “The introduced model thoroughly completes tasks like generation. The new model is more thorough than its previous model and focuses on reducing cases of ‘laziness,’ where the GPT model fails to complete a task”.

Impact of the update on user experience

User experience is important to consider, irrespective of whether in software applications. Whether you use AI for Voice Generators, code generation, or research does not matter. What matters is how you interact with an AI chatbot.

The update is a plus when it comes to user experience. You can now expect more nuanced and accurate responses, especially for code generation. Users can now enjoy more focused responses as the free moderation API identifies potentially harmful texts.

Most of the other improvements are for developers and techies. The update provides new ways for developers to understand API usage and manage API keys. As a developer, you can now assign permission keys to various API keys directly from the API keys page. A good example is when you restrict a key to access only restricted endpoints.

Developers can also turn on usage tracking for transparency. This update exposes metrics on the API key level where you can view usage on a project level, product, team, or feature. Simply have an API key for every parameter to track usage.

Industry Impact

OpenAI remains to be one of the AI industry movers. This update is likely to change the industry in several ways. First, the decreased pricing model for GPT-3.5 Turbo is likely to lower the barrier of entry for individuals and businesses looking to use OpenAI APIs. Such a move will likely increase the adoption of AI technologies in different sectors.

Users can now balance between performance and cost based on the new models. Industries like edge computing and cloud computing will likely benefit greatly from such advancements.

Developers can also enjoy increased transparency into their API usage and control how they use API keys. Regulated industries like governments, healthcare, and finance will likely benefit greatly from this update.

The embeddings are likely to enhance natural language understanding. E-commerce, content delivery systems, and customer service niches will likely experience increased engagement and user experience from this update.

We can also expect better content moderation as the update recognizes potentially harmful content. The text-moderation-007 moderation model is an updated feature to address this issue. Content-sharing online forums, websites, and social media platforms will be the biggest winners from this feature.

Ethical and Societal Considerations

Artificial intelligence is known for its impact positive impact on various industries. For instance, education is one of the areas benefiting a lot as you can use ChatGPT to Learn Data Science Faster. However, what are the ethical implications of AI advancements? What if we create models that are hard to control and regulate? Do we need to do away with teachers/ educators and adopt machines to teach us?

These are some of the questions running on most people’s minds. Introducing new API management tools is an issue of concern to developers. The company didn’t state how data sent over the API is utilized. However, OpenAI has said in the past that they don’t use data sent over the API to improve or train its models.

AI-driven chatbots may be biased on the trained data. Even though the update highlights that it will detect harmful content, it will still rely on the training data on what it understands as harmful.

Regulation is also an issue of concern in the AI age. Most governments have yet to come up with frameworks that regulate AI usage. It is still too early to determine if there are any legal implications that this new update brings.

Conclusion

OpenAI’s recent update will likely improve human interactions with AI by reducing costs, improving AI capabilities, promoting transparency & control, and increasing accessibility. The introduction of embeds promises to deal with ‘AI laziness’, mostly in code generation. The update will likely bring meaningful, seamless, and effective interactions between AI and humans in different industries.